nfs4.1 pseudo filesystem

nfs4.1 pseudo filesystem

[Ming Lei]

I want to set up the real directory as the root of the pseudo filesystem 
got exported from the server. Can I do it or I have to create /export 
empty directory as the root of the pseudo FS and then bind the real 
directories?

I have this test, both server and client running 3.11.0+:

On the server, there is /root directory.
[root@xxx ~] ls /root
anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
Desktop          disk.c  Downloads  install.log  linux3              
rpmbuild
[root@xxx ~]# ls /root/Downloads/linux-2.6.39.4
arch       crypto         firmware  Kconfig      Module.symvers  security
block      cscope.in.out  fs        kernel       net             sound
config     cscope.out     include   lib          README          tools
config-rh  cscope.po.out  init      MAINTAINERS  REPORTING-BUGS  usr
COPYING    Documentation  ipc       Makefile     samples         virt
CREDITS    drivers        Kbuild    mm           scripts

I have this /etc/exports on the server:
/root *(ro,fsid=0,crossmnt,no_root_squash)
/root/Downloads/linux-2.6.39.4 *(rw,no_root_squash,nohide)
/root/linux3/Documentation/filesystems *(rw,no_root_squash,nohide)

[root@xxx ~]# cat /proc/net/rpc/nfsd.export/content 
#path domain(flags)
# /root/.gvfs	*()
/root	*
(ro,no_root_squash,sync,wdelay,crossmnt,no_subtree_check,fsid=0,uuid=46bb42
56:d3c84b6f:bbbfff3a:8b0d912c)
[root@xxx ~]# cat /proc/net/rpc/nfsd.fh/content 
#domain fsidtype fsid [path]
* 1 0x00000000 /root


On the client side, what I saw:
[root@xxx /]# mount -o vers=4.1 10.1.13.66:/ /mnt
[root@xxx /]# cd /mnt
[root@xxx mnt]# ls
anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
Desktop          disk.c  Downloads  install.log  linux3              
rpmbuild
[root@xxx mnt]# cd Downloads/linux-2.6.39.4
[root@xxx linux-2.6.39.4]# ls
arch       COPYING        cscope.out     firmware  ipc      lib          
Module.symvers  samples   tools
block      CREDITS        cscope.po.out  fs        Kbuild   MAINTAINERS  
net             scripts   usr
config     crypto         Documentation  include   Kconfig  Makefile     
README          security  virt
config-rh  cscope.in.out  drivers        init      kernel   mm           
REPORTING-BUGS  sound
[root@xxx linux-2.6.39.4]# rm REPORTING-BUGS 
rm: remove regular file `REPORTING-BUGS'? y
rm: cannot remove `REPORTING-BUGS': Read-only file system

The question is:
Why can I delete this file even I specify the 
mountpoint /root/Downloads/linux-2.6.39.4 to allow RW access?

Thanks
Ming

Re: nfs4.1 pseudo filesystem

[J. Bruce Fields]

On Thu, Oct 10, 2013 at 09:39:38PM +0000, Ming Lei wrote:
> 
> I want to set up the real directory as the root of the pseudo filesystem 
> got exported from the server.

Unless you have some very good reason, I recommend just ignoring the
pseudo filesystem.  Just export the directories you want to without
using fsid=0 or bind mounts, just as you would with NFSv2/v3.  That will
work fine on any recent server.

--b.

> Can I do it or I have to create /export 
> empty directory as the root of the pseudo FS and then bind the real 
> directories?
> 
> I have this test, both server and client running 3.11.0+:
> 
> On the server, there is /root directory.
> [root@xxx ~] ls /root
> anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> Desktop          disk.c  Downloads  install.log  linux3              
> rpmbuild
> [root@xxx ~]# ls /root/Downloads/linux-2.6.39.4
> arch       crypto         firmware  Kconfig      Module.symvers  security
> block      cscope.in.out  fs        kernel       net             sound
> config     cscope.out     include   lib          README          tools
> config-rh  cscope.po.out  init      MAINTAINERS  REPORTING-BUGS  usr
> COPYING    Documentation  ipc       Makefile     samples         virt
> CREDITS    drivers        Kbuild    mm           scripts
> 
> I have this /etc/exports on the server:
> /root *(ro,fsid=0,crossmnt,no_root_squash)
> /root/Downloads/linux-2.6.39.4 *(rw,no_root_squash,nohide)
> /root/linux3/Documentation/filesystems *(rw,no_root_squash,nohide)
> 
> [root@xxx ~]# cat /proc/net/rpc/nfsd.export/content 
> #path domain(flags)
> # /root/.gvfs	*()
> /root	*
> (ro,no_root_squash,sync,wdelay,crossmnt,no_subtree_check,fsid=0,uuid=46bb42
> 56:d3c84b6f:bbbfff3a:8b0d912c)
> [root@xxx ~]# cat /proc/net/rpc/nfsd.fh/content 
> #domain fsidtype fsid [path]
> * 1 0x00000000 /root
> 
> 
> On the client side, what I saw:
> [root@xxx /]# mount -o vers=4.1 10.1.13.66:/ /mnt
> [root@xxx /]# cd /mnt
> [root@xxx mnt]# ls
> anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> Desktop          disk.c  Downloads  install.log  linux3              
> rpmbuild
> [root@xxx mnt]# cd Downloads/linux-2.6.39.4
> [root@xxx linux-2.6.39.4]# ls
> arch       COPYING        cscope.out     firmware  ipc      lib          
> Module.symvers  samples   tools
> block      CREDITS        cscope.po.out  fs        Kbuild   MAINTAINERS  
> net             scripts   usr
> config     crypto         Documentation  include   Kconfig  Makefile     
> README          security  virt
> config-rh  cscope.in.out  drivers        init      kernel   mm           
> REPORTING-BUGS  sound
> [root@xxx linux-2.6.39.4]# rm REPORTING-BUGS 
> rm: remove regular file `REPORTING-BUGS'? y
> rm: cannot remove `REPORTING-BUGS': Read-only file system
> 
> The question is:
> Why can I delete this file even I specify the 
> mountpoint /root/Downloads/linux-2.6.39.4 to allow RW access?
> 
> Thanks
> Ming
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: nfs4.1 pseudo filesystem

[Ming Lei]

I want to cherry-pick a few subdirectories under say /root, but I don't want to specifies all these mount directories one by one just like NFSv2/3.

Does pseudo filesystem in kernel v3.11 work well enough?

Thanks
Ming

-----Original Message-----
From: J. Bruce Fields [mailto:bfields@fieldses.org] 
Sent: Friday, October 11, 2013 9:30 AM
To: Ming Lei
Cc: linux-nfs@vger.kernel.org
Subject: Re: nfs4.1 pseudo filesystem

On Thu, Oct 10, 2013 at 09:39:38PM +0000, Ming Lei wrote:
> 
> I want to set up the real directory as the root of the pseudo 
> filesystem got exported from the server.

Unless you have some very good reason, I recommend just ignoring the pseudo filesystem.  Just export the directories you want to without using fsid=0 or bind mounts, just as you would with NFSv2/v3.  That will work fine on any recent server.

--b.

> Can I do it or I have to create /export empty directory as the root of 
> the pseudo FS and then bind the real directories?
> 
> I have this test, both server and client running 3.11.0+:
> 
> On the server, there is /root directory.
> [root@xxx ~] ls /root
> anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> Desktop          disk.c  Downloads  install.log  linux3              
> rpmbuild
> [root@xxx ~]# ls /root/Downloads/linux-2.6.39.4
> arch       crypto         firmware  Kconfig      Module.symvers  security
> block      cscope.in.out  fs        kernel       net             sound
> config     cscope.out     include   lib          README          tools
> config-rh  cscope.po.out  init      MAINTAINERS  REPORTING-BUGS  usr
> COPYING    Documentation  ipc       Makefile     samples         virt
> CREDITS    drivers        Kbuild    mm           scripts
> 
> I have this /etc/exports on the server:
> /root *(ro,fsid=0,crossmnt,no_root_squash)
> /root/Downloads/linux-2.6.39.4 *(rw,no_root_squash,nohide) 
> /root/linux3/Documentation/filesystems *(rw,no_root_squash,nohide)
> 
> [root@xxx ~]# cat /proc/net/rpc/nfsd.export/content #path 
> domain(flags)
> # /root/.gvfs	*()
> /root	*
> (ro,no_root_squash,sync,wdelay,crossmnt,no_subtree_check,fsid=0,uuid=4
> 6bb42
> 56:d3c84b6f:bbbfff3a:8b0d912c)
> [root@xxx ~]# cat /proc/net/rpc/nfsd.fh/content #domain fsidtype fsid 
> [path]
> * 1 0x00000000 /root
> 
> 
> On the client side, what I saw:
> [root@xxx /]# mount -o vers=4.1 10.1.13.66:/ /mnt [root@xxx /]# cd 
> /mnt [root@xxx mnt]# ls
> anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> Desktop          disk.c  Downloads  install.log  linux3              
> rpmbuild
> [root@xxx mnt]# cd Downloads/linux-2.6.39.4 [root@xxx linux-2.6.39.4]# 
> ls
> arch       COPYING        cscope.out     firmware  ipc      lib          
> Module.symvers  samples   tools
> block      CREDITS        cscope.po.out  fs        Kbuild   MAINTAINERS  
> net             scripts   usr
> config     crypto         Documentation  include   Kconfig  Makefile     
> README          security  virt
> config-rh  cscope.in.out  drivers        init      kernel   mm           
> REPORTING-BUGS  sound
> [root@xxx linux-2.6.39.4]# rm REPORTING-BUGS
> rm: remove regular file `REPORTING-BUGS'? y
> rm: cannot remove `REPORTING-BUGS': Read-only file system
> 
> The question is:
> Why can I delete this file even I specify the mountpoint 
> /root/Downloads/linux-2.6.39.4 to allow RW access?
> 
> Thanks
> Ming
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" 
> in the body of a message to majordomo@vger.kernel.org More majordomo 
> info at  http://vger.kernel.org/majordomo-info.html

Re: nfs4.1 pseudo filesystem

[J. Bruce Fields]

On Fri, Oct 11, 2013 at 04:51:29PM +0000, Ming Lei wrote:
> I want to cherry-pick a few subdirectories under say /root, but I don't want to specifies all these mount directories one by one just like NFSv2/3.

Well, you're kind of stuck enumerating them whether as exports or bind
mounts.

> Does pseudo filesystem in kernel v3.11 work well enough?

But it should work anyway, yes.

Oh, wait, I see the problem: it doesn't work well if you're exporting
subdirectories and not full filesystems.  Unless you do something
special at the export points  (e.g. mount --bind them over themselves),
knfsd doesn't know when you've crossed an export point.

--b.

> 
> Thanks
> Ming
> 
> -----Original Message-----
> From: J. Bruce Fields [mailto:bfields@fieldses.org] 
> Sent: Friday, October 11, 2013 9:30 AM
> To: Ming Lei
> Cc: linux-nfs@vger.kernel.org
> Subject: Re: nfs4.1 pseudo filesystem
> 
> On Thu, Oct 10, 2013 at 09:39:38PM +0000, Ming Lei wrote:
> > 
> > I want to set up the real directory as the root of the pseudo 
> > filesystem got exported from the server.
> 
> Unless you have some very good reason, I recommend just ignoring the pseudo filesystem.  Just export the directories you want to without using fsid=0 or bind mounts, just as you would with NFSv2/v3.  That will work fine on any recent server.
> 
> --b.
> 
> > Can I do it or I have to create /export empty directory as the root of 
> > the pseudo FS and then bind the real directories?
> > 
> > I have this test, both server and client running 3.11.0+:
> > 
> > On the server, there is /root directory.
> > [root@xxx ~] ls /root
> > anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> > Desktop          disk.c  Downloads  install.log  linux3              
> > rpmbuild
> > [root@xxx ~]# ls /root/Downloads/linux-2.6.39.4
> > arch       crypto         firmware  Kconfig      Module.symvers  security
> > block      cscope.in.out  fs        kernel       net             sound
> > config     cscope.out     include   lib          README          tools
> > config-rh  cscope.po.out  init      MAINTAINERS  REPORTING-BUGS  usr
> > COPYING    Documentation  ipc       Makefile     samples         virt
> > CREDITS    drivers        Kbuild    mm           scripts
> > 
> > I have this /etc/exports on the server:
> > /root *(ro,fsid=0,crossmnt,no_root_squash)
> > /root/Downloads/linux-2.6.39.4 *(rw,no_root_squash,nohide) 
> > /root/linux3/Documentation/filesystems *(rw,no_root_squash,nohide)
> > 
> > [root@xxx ~]# cat /proc/net/rpc/nfsd.export/content #path 
> > domain(flags)
> > # /root/.gvfs	*()
> > /root	*
> > (ro,no_root_squash,sync,wdelay,crossmnt,no_subtree_check,fsid=0,uuid=4
> > 6bb42
> > 56:d3c84b6f:bbbfff3a:8b0d912c)
> > [root@xxx ~]# cat /proc/net/rpc/nfsd.fh/content #domain fsidtype fsid 
> > [path]
> > * 1 0x00000000 /root
> > 
> > 
> > On the client side, what I saw:
> > [root@xxx /]# mount -o vers=4.1 10.1.13.66:/ /mnt [root@xxx /]# cd 
> > /mnt [root@xxx mnt]# ls
> > anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
> > Desktop          disk.c  Downloads  install.log  linux3              
> > rpmbuild
> > [root@xxx mnt]# cd Downloads/linux-2.6.39.4 [root@xxx linux-2.6.39.4]# 
> > ls
> > arch       COPYING        cscope.out     firmware  ipc      lib          
> > Module.symvers  samples   tools
> > block      CREDITS        cscope.po.out  fs        Kbuild   MAINTAINERS  
> > net             scripts   usr
> > config     crypto         Documentation  include   Kconfig  Makefile     
> > README          security  virt
> > config-rh  cscope.in.out  drivers        init      kernel   mm           
> > REPORTING-BUGS  sound
> > [root@xxx linux-2.6.39.4]# rm REPORTING-BUGS
> > rm: remove regular file `REPORTING-BUGS'? y
> > rm: cannot remove `REPORTING-BUGS': Read-only file system
> > 
> > The question is:
> > Why can I delete this file even I specify the mountpoint 
> > /root/Downloads/linux-2.6.39.4 to allow RW access?
> > 
> > Thanks
> > Ming
> > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-nfs" 
> > in the body of a message to majordomo@vger.kernel.org More majordomo 
> > info at  http://vger.kernel.org/majordomo-info.html

Re: nfs4.1 pseudo filesystem

[J. Bruce Fields]

On Fri, Oct 11, 2013 at 02:29:42PM -0400, J. Bruce Fields wrote:
> On Fri, Oct 11, 2013 at 04:51:29PM +0000, Ming Lei wrote:
> > I want to cherry-pick a few subdirectories under say /root, but I don't want to specifies all these mount directories one by one just like NFSv2/3.
> 
> Well, you're kind of stuck enumerating them whether as exports or bind
> mounts.
> 
> > Does pseudo filesystem in kernel v3.11 work well enough?
> 
> But it should work anyway, yes.
> 
> Oh, wait, I see the problem: it doesn't work well if you're exporting
> subdirectories and not full filesystems.  Unless you do something
> special at the export points  (e.g. mount --bind them over themselves),
> knfsd doesn't know when you've crossed an export point.

I wonder if exportfs should be taught to mark export points as
junctions.  That might fix this kind of problem?

It might mean creating junctions never get cleaned up, which is ugly
though perhaps unlikely to be a serious problem in practice.

--b.

Re: nfs4.1 pseudo filesystem

[Chuck Lever]

On Oct 11, 2013, at 4:58 PM, "J. Bruce Fields" <bfields@fieldses.org> wrote:

> On Fri, Oct 11, 2013 at 02:29:42PM -0400, J. Bruce Fields wrote:
>> On Fri, Oct 11, 2013 at 04:51:29PM +0000, Ming Lei wrote:
>>> I want to cherry-pick a few subdirectories under say /root, but I don't want to specifies all these mount directories one by one just like NFSv2/3.
>> 
>> Well, you're kind of stuck enumerating them whether as exports or bind
>> mounts.
>> 
>>> Does pseudo filesystem in kernel v3.11 work well enough?
>> 
>> But it should work anyway, yes.
>> 
>> Oh, wait, I see the problem: it doesn't work well if you're exporting
>> subdirectories and not full filesystems.  Unless you do something
>> special at the export points  (e.g. mount --bind them over themselves),
>> knfsd doesn't know when you've crossed an export point.
> 
> I wonder if exportfs should be taught to mark export points as
> junctions.  That might fix this kind of problem?

A junction is always an unreadable directory.

We'd be better off with a generic "please upcall" bit for directories.  Then mountd could sort out what are junctions and what are legitimate exports, and the kernel would leave all the other directories in an export alone.

> It might mean creating junctions never get cleaned up, which is ugly
> though perhaps unlikely to be a serious problem in practice.

-- 
Chuck Lever
chuck[dot]lever[at]oracle[dot]com

RE: nfs4.1 pseudo filesystem

[Wangminlan]

Hi, Ming,
	I guess the problem here is that the filesystem on "/root", "/root/Downloads/linux-2.6.39.4", and "/root/linux3/Documentation/filesystems" are the same filesystem.
	So when your client tried to access anything within the three directories, they all use the export "/root *(ro,fsid=0,crossmnt,no_root_squash)" to determin their permission.
	Try to mount some other filesystems on your "real" exports, I mean, the exports with fsid other than 0, then do your tests.

B.R
Minlan Wang


-----Original Message-----
From: linux-nfs-owner@vger.kernel.org [mailto:linux-nfs-owner@vger.kernel.org] On Behalf Of Ming Lei
Sent: Friday, October 11, 2013 5:40 AM
To: linux-nfs@vger.kernel.org
Subject: nfs4.1 pseudo filesystem


I want to set up the real directory as the root of the pseudo filesystem 
got exported from the server. Can I do it or I have to create /export 
empty directory as the root of the pseudo FS and then bind the real 
directories?

I have this test, both server and client running 3.11.0+:

On the server, there is /root directory.
[root@xxx ~] ls /root
anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
Desktop          disk.c  Downloads  install.log  linux3              
rpmbuild
[root@xxx ~]# ls /root/Downloads/linux-2.6.39.4
arch       crypto         firmware  Kconfig      Module.symvers  security
block      cscope.in.out  fs        kernel       net             sound
config     cscope.out     include   lib          README          tools
config-rh  cscope.po.out  init      MAINTAINERS  REPORTING-BUGS  usr
COPYING    Documentation  ipc       Makefile     samples         virt
CREDITS    drivers        Kbuild    mm           scripts

I have this /etc/exports on the server:
/root *(ro,fsid=0,crossmnt,no_root_squash)
/root/Downloads/linux-2.6.39.4 *(rw,no_root_squash,nohide)
/root/linux3/Documentation/filesystems *(rw,no_root_squash,nohide)

[root@xxx ~]# cat /proc/net/rpc/nfsd.export/content 
#path domain(flags)
# /root/.gvfs	*()
/root	*
(ro,no_root_squash,sync,wdelay,crossmnt,no_subtree_check,fsid=0,uuid=46bb42
56:d3c84b6f:bbbfff3a:8b0d912c)
[root@xxx ~]# cat /proc/net/rpc/nfsd.fh/content 
#domain fsidtype fsid [path]
* 1 0x00000000 /root


On the client side, what I saw:
[root@xxx /]# mount -o vers=4.1 10.1.13.66:/ /mnt
[root@xxx /]# cd /mnt
[root@xxx mnt]# ls
anaconda-ks.cfg  disk    Documents  fs_log.old   install.log.syslog  Public
Desktop          disk.c  Downloads  install.log  linux3              
rpmbuild
[root@xxx mnt]# cd Downloads/linux-2.6.39.4
[root@xxx linux-2.6.39.4]# ls
arch       COPYING        cscope.out     firmware  ipc      lib          
Module.symvers  samples   tools
block      CREDITS        cscope.po.out  fs        Kbuild   MAINTAINERS  
net             scripts   usr
config     crypto         Documentation  include   Kconfig  Makefile     
README          security  virt
config-rh  cscope.in.out  drivers        init      kernel   mm           
REPORTING-BUGS  sound
[root@xxx linux-2.6.39.4]# rm REPORTING-BUGS 
rm: remove regular file `REPORTING-BUGS'? y
rm: cannot remove `REPORTING-BUGS': Read-only file system

The question is:
Why can I delete this file even I specify the 
mountpoint /root/Downloads/linux-2.6.39.4 to allow RW access?

Thanks
Ming

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html